Highlights
- Evaluates multiple LLMs across simulated personas
- Dashboard + command-driven console MVPs
- Structured scoring and analytics for product messaging
- Full AWS deployment with secure delivery
- Backend architecture built around LLM prompt orchestration
Summary
Nextmost evaluates how different Large Language Models describe, position, and recommend products for simulated customer personas. It replaces intuition-based messaging decisions with analytics and structured scoring — guiding strategy and identifying which models perform best in specific contexts.
Client
Armando Kirwin, creative technologist with projects showcased at Cannes Film Festival, Venice Biennale, and SXSW. Co-founder of Artie (acquired 2024).
Scope & Key Deliverables
- Designed and developed two MVP versions to validate core capabilities
- Dashboard (v1) — Interactive analytics for persona-based evaluation
- Console Interface (v2) — Command-driven web app for rapid simulation
- Backend architecture for prompt orchestration and scoring
- Structured PostgreSQL schema for analytics and comparison
- Secure authentication and environment-based configuration
- Full AWS deployment with HTTPS, logging and automation
Tech Stack
- Frontend (v1): Next.js, Tremor UI, ECharts, Chart.js
- Frontend (v2): Static console web client (HTML/CSS/JS)
- Backend: FastAPI, Uvicorn, OpenRouter API
- Database: PostgreSQL (structured analytics schema)
- Infra: AWS EC2 + Nginx + Certbot (SSL) + systemd
Technical Overview
Nextmost unifies LLM orchestration and analytics visualization into a single simulation platform. FastAPI manages persona definitions, prompt templates, and multi-model requests via OpenRouter. Simulation outputs flow into PostgreSQL, powering visual comparisons and strategy insights.
Two complementary frontends connect to the same API — a Next.js dashboard for visual analysis, and a lightweight console for rapid experimentation. Deployed over AWS EC2 with secure HTTPS, automated services, access control, and monitoring — ensuring stability during R&D scaling.